Testimony to the U.S. Senate Committee on Homeland Security and Governmental Affairs, Subcommittee on Disaster Management, District of Columbia, and Census, on AI agent incidents

Chris Painter's testimony, posted on METR's blog, to a U.S. Senate subcommittee on AI agent incidents. The companion cites it in Box 3 for the setting of the July 2026 breach, evaluations run with safeguards reduced and no monitor reading the agents' reasoning, and for the agents re-establishing their erased board and validating a method of cheating the grader. It supports the account of organization without metabolism.

Cited in

the companion paper (Companion Box 3)

Used for

Mesocosm breach; Box 3

Strand

Agents in the wild: incidents & field evidence · The July 2026 OpenAI-Hugging Face incident

Cite

@techreport{metr2026testimony,
  title = {Testimony to the U.S. Senate Committee on Homeland Security and Governmental Affairs, Subcommittee on Disaster Management, District of Columbia, and Census, on AI agent incidents},
  author = {Chris Painter},
  year = {2026},
  howpublished = {METR blog},
  url = {https://metr.org/blog/2026-09-30-chris-painter-senate-testimony/},
}